An AI salesperson for 265,000 products

A major online retailer of appliances and electronics asked us to improve its contact centre operations. We deployed Taron, our AI salesperson, to advise customers across the full product range, help them choose and place orders in chat. Customers receive replies more than six times faster on average, while the low cost of handling conversations makes it possible to offer personal advice to many more shoppers.

A good salesperson needs to know the whole range

The retailer sells everything from refrigerators and dishwashers to electronics, kitchenware and garden products. Shoppers often struggle to compare specifications and choose a model, so they turn to the website chat or a messaging app. A knowledgeable salesperson can explain the differences, address concerns and help someone buy when they might otherwise leave.

Good advice makes a purchase more likely, so it makes sense to invite more shoppers into a conversation. But every chat takes staff time, and more traffic means more people. Round-the-clock coverage also requires expertise across categories on every shift. There may be few enquiries at night, but nobody knows in advance what those customers will need. One person is unlikely to cover everything, while a full team is expensive.

When we reviewed actual conversations, we found another problem: the average response took more than three minutes. That is a long wait for someone choosing a product right now.

Adapting Taron to the retailer’s catalogue and sales process

Taron is built for large catalogues. It indexes products with all available attributes and gives language models tools to search and compare them. During deployment, we connected the entire range and adapted the system to the retailer’s data. Different categories have different properties and buying criteria, and the store’s category tree is not always the best structure for a consultation. Taron reorganises it for the large language model (LLM), while preserving the mapping to the original catalogue.

Sales instructions needed separate work. Scripts written for people cannot simply be passed to a model. We analysed real conversations and worked with the client on guidance for each category: what to ask, how to compare options, explain differences and respond to objections.

A good salesperson also understands the store’s commercial priorities. Taron can consider product margins, recommend accessories, and offer extended warranties or other services where appropriate. The influence of margins on recommendations is adjustable and can be switched off entirely.

From the first question to an order

We connected Taron to the retailer’s existing platform for website and messaging enquiries. Customers use their usual channel while the AI clarifies their needs, selects products, explains trade-offs and suggests relevant extras. It can collect the necessary details, agree on delivery or collection, and place the order in the retailer’s internal system directly from the chat.

When a situation needs a person’s decision, such as a complaint or a request for special terms, Taron hands the conversation to an operator in the same platform. If a customer stops replying in a messaging app, it follows up gently after about an hour and offers to continue, helping the store recover unfinished conversations.

Personal advice for more shoppers

With Taron, customers receive replies more than six times faster on average, and no reply takes more than five minutes. We estimate that handling a message costs around one tenth as much as it would in a remote contact centre.

This changes how the retailer can organise its sales funnel. With human operators, more enquiries have to be balanced against staffing costs, shifts and workload. With Taron, the store can direct much more traffic into chat because conversations are inexpensive to handle and easy to scale.

The more shoppers receive useful advice, the more opportunities the store has to turn interest into an order.

The retailer can offer help with choosing instead of leaving shoppers alone with a huge catalogue. Customers get a clearer understanding of the options; the business has a way to improve conversion and generate more orders from incoming traffic. Access to that advice depends far less on which specialists happen to be on shift.